Sharma P.,University of Southern California |
Huang C.,NEC Labs |
Nevatia R.,University of Southern California
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition | Year: 2012
Most common approaches for object detection collect thousands of training examples and train a detector in an offline setting, using supervised learning methods, with the objective of obtaining a generalized detector that would give good performance on various test datasets. However, when an offline trained detector is applied on challenging test datasets, it may fail in some cases by not being able to detect some objects or by producing false alarms. We propose an unsupervised multiple instance learning (MIL) based incremental solution to deal with this issue. We introduce an MIL loss function for Real Adaboost and present a tracking based effective unsupervised online sample collection mechanism to collect the online samples for incremental learning. Experiments demonstrate the effectiveness of our approach by improving the performance of a state of the art offline trained detector on the challenging datasets for pedestrian category. © 2012 IEEE.
Wang Z.,Princeton University |
Fok M.P.,Princeton University |
Xu L.,NEC Labs |
Chang J.,Princeton University |
Prucnal P.R.,Princeton University
Optics Express | Year: 2010
Temporal phase modulation of spread stealth signals is proposed and demonstrated to improve optical steganography transmission privacy. After phase modulation, the temporally spread stealth signal has a more complex spectral-phase-temporal relationship, such that the original temporal profile cannot be restored when only dispersion compensation is applied to the temporally spread stealth signals. Therefore, it increases the difficulty for the eavesdropper to detect and intercept the stealth channel that is hidden under a public transmission, even with a correct dispersion compensation device. The experimental results demonstrate the feasibility of this approach and display insignificant degradation in transmission performance, compared to the conventional stealth transmission without temporal phase modulation. The proposed system can also work without a clock transmission for signal synchronization. Our analysis and simulation results show that it is difficult for the adversary to detect the existence of the stealth transmission, or find the correct phase mask to recover the stealth signals. © 2010 Optical Society of America.
Kahlon V.,NEC Labs
2012 Formal Methods in Computer-Aided Design, FMCAD 2012 | Year: 2012
Triggering errors in concurrent programs is a notoriously difficult task. A key reason for this is the behavioral complexity resulting from the large number of interleavings of operations of different threads. An even more challenging task is fixing errors once they are detected. In general, automatically synthesizing a correct program from a buggy one is a hard problem. However for simple correctness properties that depend on the syntactic structure of the program rather than its semantics, automatic error correction becomes feasible. In this paper, we consider the problem of lock insertion to enforce critical sections required to fix bugs like atomicity violations. A key challenge in lock insertion is that enforcing critical sections is not the sole criterion that needs to be satisfied. Often other correctness constraints like deadlock-freedom also need to be met. Moreover, apart from ensuring correctness, another key concern during lock insertion is performance. Indeed, mutual exclusion constraints generated by locks kill parallelism thereby impacting performance. Thus it is crucial that the newly introduced critical sections be kept as small as possible. In other words, our goal is lock insertion while meeting the dual, and often conflicting, requirements of (i) correctness and (ii) performance. In this paper, we present a fully automatic, provable optimal, efficient and precise technique for lock insertion in concurrent code that ensures deadlock freedom while attempting to minimize the resulting critical sections. © 2012 IEEE.
Liu N.N.,Hong Kong University of Science and Technology |
He L.,Hong Kong University of Science and Technology |
Zhao M.,NEC Labs
ACM Transactions on Intelligent Systems and Technology | Year: 2013
Most existing collaborative filtering models only consider the use of user feedback (e.g., ratings) and meta data (e.g., content, demographics). However, in most real world recommender systems, context information, such as time and social networks, are also very important factors that could be considered in order to produce more accurate recommendations. In this work, we address several challenges for the context aware movie recommendation tasks in CAMRa 2010: (1) how to combine multiple heterogeneous forms of user feedback? (2) how to cope with dynamic user and item characteristics? (3) how to capture and utilize social connections among users? For the first challenge, we propose a novel ranking based matrix factorization model to aggregate explicit and implicit user feedback. For the second challenge, we extend this model to a sequential matrix factorization model to enable time-aware parametrization. Finally, we introduce a network regularization function to constrain user parameters based on social connections. To the best of our knowledge, this is the first study that investigates the collective modeling of social and temporal dynamics. Experiments on the CAMRa 2010 dataset demonstrated clear improvements over many baselines. © 2013 ACM.
Kahlon V.,NEC Labs
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2012
The key to making program analysis practical for large concurrent programs is to isolate a small set of interleavings to be explored without losing precision of the analysis at hand. The state-of-the-art in restricting the set of interleavings while guaranteeing soundness is partial order reduction (POR). The main idea behind POR is to partition all interleavings of the given program into equivalence classes based on the partial orders they induce on shared objects. Then for each partial order at least one interleaving need be explored. POR classifies two interleavings as non-equivalent if executing them leads to different values of shared variables. However, some of the most common properties about concurrent programs like detection of data races, deadlocks and atomicity as well as assertion violations reduce to control state reachability. We exploit the key observation that even though different interleavings may lead to different values of program variables, they may induce the same control behavior. Hence these interleavings, which induce different partial orders, can in fact be treated as being equivalent. Since in most concurrent programs threads are loosely coupled, i.e., the values of shared variables typically flow into a small number of conditional statements of threads, we show that classifying interleavings based on the control behaviors rather than the partial orders they induce, drastically reduces the number of interleavings that need be explored. In order to exploit this loose coupling we leverage the use of dataflow analysis for concurrent programs, specifically numerical domains. This, in turn, greatly enhances the scalability of concurrent program analysis. © 2012 Springer-Verlag Berlin Heidelberg.